As for the header/implementation split, I might have been brainwashed by the language, but I like it :) I take it as a nice division between the public interfac

As for the header/implementation split, I might have been brainwashed by the language, but I like it :) I take it as a nice division between the public interface and private implementation. And now that we have purely synthesized properties in the modern runtimes, we can keep the header file really clean: https://gist.github.com/766381 https://gist.github.com/766381 I have never missed allocating objects on the stack, I am mostly happy with autorelease pools. They are slower and “less deterministic” than stack-based objects, but I don’t mind.